Horton
Menu

Re: Deleting older versions in unique indexes to avoid page splits

Simon Riggs <simon@2ndquadrant.com>

View thread
From: Simon Riggs <simon@2ndquadrant.com>
To: Peter Geoghegan <pg@bowt.ie>
Cc: Anastasia Lubennikova <a.lubennikova@postgrespro.ru>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2020-10-24T09:55:13Z
Lists: pgsql-hackers 
On Fri, 23 Oct 2020 at 18:14, Peter Geoghegan <pg@bowt.ie> wrote:

> It's obvious that a page split is more expensive than the delete
> operation (when it works out).

The problem I highlighted is that the average UPDATE latency is x2
what it is on current HEAD. That is not consistent with the reported
TPS, so it remains an issue and that isn't obvious.

> It doesn't need a microbenchmark (and I
> really can't think of one that would make any sense).

I'm asking for detailed timings so we can understand the latency
issue. I didn't ask for a microbenchmark.

I celebrate your results, but we do need to understand the issue, somehow.

-- 
Simon Riggs                http://www.EnterpriseDB.com/

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Enhance nbtree index tuple deletion.

  2. Pass down "logically unchanged index" hint.

  3. Fix index deletion latestRemovedXid bug.

  4. Deprecate nbtree's BTP_HAS_GARBAGE flag.